Coordinating Search-Informed Reasoning and Reasoning-Guided Search in Claim Verification

Qisheng Hu; Quanyu Long; Wenya Wang

REVIEW 4 major objections 5 minor 47 references

Coordinating Search-Informed Reasoning and Reasoning-Guided Search in Claim Verification

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Two cooperating agents—one reasoning, one searching—beat single-model pipelines on multi-hop claim verification.

desk verdict A real two-agent RL design for claim verification, but the empirical headline is not yet supported by the reported numbers. read the letter →

arxiv 2506.07528 v2 pith:CXD3EBSE submitted 2025-06-09 cs.AI

Qisheng Hu , Quanyu Long , Wenya Wang This is my paper

classification cs.AI

keywords multi-hopclaimverificationhierarchicalagentsreinforcementlearningGRPOretrieval-augmentedreasoningfact-checkingLLMcoordinatedtraining

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

HARIS claims that multi-hop claim verification is best done by splitting the job between a high-level reasoning agent and a low-level search agent that trade questions and evidence back and forth, rather than having one LLM do all retrieval and reasoning in a single pass. The paper argues that this division lets each agent specialize, with the reasoning agent deciding what is missing and the search agent iteratively refining queries through a Wikipedia corpus. Both agents are trained together with GRPO, an outcome-reward reinforcement learning method, so the collaboration itself is optimized rather than hand-coded. If the claim holds, a relatively small 4-billion-parameter model can beat much larger or more elaborate verification pipelines on the EX-FEVER, HOVER, and CHECKWHY benchmarks, especially on the hardest 3-hop and 4-hop claims.

What carries the argument

The load-bearing object is the two-agent coordination loop trained by Group Relative Policy Optimization (GRPO), an RL variant that normalizes rewards within groups and uses a KL penalty to keep the policy stable. Specialized tags (<question>, <search>, <result>, <report>, <verification>) create a structured protocol between the two agents, and retrieved tokens are masked from the policy-gradient loss so the model learns from its own generation. The alternating training schedule—foundation stage followed by segment-wise alternation—is what keeps the search agent aligned with the questions the reasoning agent actually asks.

What would settle it

Take HARIS's trained reasoning agent and swap in a search agent trained on human-annotated question-answer pairs instead of GPT-4o pseudo-answers, keeping everything else fixed; if verification accuracy does not drop, the synthesized training signal is not what makes the method work. Alternatively, run the same coordinated training with the LLM-as-a-Judge reward replaced by exact-match F1 on gold answers: the paper's own comparison predicts a roughly 3-point average loss, so a smaller or absent drop would falsify the claim that the judge reward drives search quality.

Watch

Extended reading notes

Core claim

The central discovery is that the interleaving of reasoning and retrieval—reasoning-guided search and search-informed reasoning—can be explicitly modeled as a two-level agent system and trained end-to-end. The high-level reasoning agent constructs the verification chain and emits factual questions in <question> tags; the low-level search agent executes queries, reads top-3 retrieved passages, and reports evidence in <report> tags. Training alternates: questions sampled from the reasoning agent's rollouts, paired with GPT-4o pseudo-answers derived from dataset evidence, train the search agent, and the updated search agent then supports another round of reasoning-agent training. Across five hop settings on HOVER and EX-FEVER plus a transfer test on CHECKWHY, HARIS reports the highest F1 and accuracy among RAG, decompose-then-verify, program-guided, FOL, and search-augmented baselines, with its largest margins on 3-hop and 4-hop verification.

Load-bearing premise

The whole training loop depends on the belief that questions sampled from the reasoning agent's own early rollouts, answered by GPT-4o pseudo-ground-truth built from dataset evidence, are good enough to teach the search agent what to retrieve; if the sampled questions are off-target or the pseudo-answers are noisy, both agents learn the wrong collaboration.

Editorial extensions

If this is right

Harder multi-hop claims (3-hop and 4-hop) stand to gain the most from role separation, since HARIS's largest margins over baselines appear at those depths.
A 4B-parameter model trained this way can outperform a 32B reasoning model (Search-o1's QwQ-32B-preview) on the same verification task, suggesting that architecture and training signal matter more than raw scale.
LLM-as-a-Judge rewards for the search agent produce more informative evidence-gathering than exact-match rewards, leading to fewer search calls and higher final verification reward.
Because the interaction uses explicit question and report tags, the verification process is inspectable step by step, making the system's evidence chain interpretable.
With the same 4B base, HARIS beats supervised finetuning on every dataset and hop setting, implying that the RL coordination, not just imitation of correct answers, carries the improvement.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Beyond the paper: the same alternating question-synthesis loop should transfer to open-domain QA and multi-hop retrieval tasks, where the reward would need to be answer correctness rather than claim verdict.
Beyond the paper: because the search agent is trained on questions the reasoning agent asks, the framework should be sensitive to distribution drift in claims; a test set with different styles of claims would reveal how much of the gain is task-specific.
Beyond the paper: one could test the separation hypothesis directly by replacing the learned search agent with a fixed, non-trained retrieval policy (for example, retrieving the claim's noun phrases); if HARIS still wins, the search-agent training is not the active ingredient.
Beyond the paper: the reported Cohen's kappa of 0.81 validates the LLM judge on 150 sampled questions; scaling that human check to several hundred across adversarial, partially-supported claims would show whether the reward remains aligned on the cases the system gets wrong.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Short version: HARIS is a genuine new design—two specialized agents, one for reasoning and one for search, trained jointly with GRPO—and the paper is worth engaging with. The ablation story is coherent, the code is out, the human eval of the judge is a plus, and evaluation is on external benchmarks with held-out test subsets. But the 'consistently outperforms' claim is weaker than the prose. The HOVER margins over the best baseline are a couple of F1 points, which is inside the sampling noise for 500-instance sets, and the one same-backbone table (Table 5) has a FOLK row identical to HARIS, which looks like a copy-paste error. The paper deserves a serious referee, but the headline needs more evidence before it's convincing.

New: I don't know of a prior claim-verification paper that splits reasoning and search into two agents and trains both with outcome-based RL in an alternating schedule. The ablations support the split (single-agent vs. HARIS; LLM-as-judge vs. F1 reward), the coordination-round study is informative even if the result is not monotonic, and the related work covers the recent search-RL literature (Search-o1, Search-R1, R1-Searcher). The citation pattern looks appropriate.

Soft spots. Table 5 is the main problem: as printed, the direct 4B-vs-4B comparison is unusable because the FOLK row is identical to HARIS on all five settings. That is a concrete error, not an interpretation issue. The main tables have no confidence intervals, no multiple seeds, and no significance tests. On a 500-sample binary accuracy near 0.6, the SE of a difference between two methods is about 3.1 points, so a 95% interval spans roughly ±6 points; the HOVER gains over QACheck are +1.8, +2.2, and +2.0 accuracy points, all within that noise. The consistency across the six settings is suggestive but isn't a tested joint claim. Also, the claim that increasing coordination rounds N 'generally leads to stronger performance' is not fully supported by Table 3: N=3 hurts HOVER-4hop and ties HOVER-2hop. The GPT-4o pseudo-answer pipeline for search-agent training is a reasonable way to bootstrap a training signal, but it is a self-referential loop; the human eval covers the judge, not the pseudo-answers.

Net: the architecture is a solid contribution to the fact-checking and agentic-RAG subfields, and the paper should go to peer review. I'd condition acceptance on fixing Table 5 and adding variance-aware reporting. The size of the improvement over strong baselines is plausible but not yet demonstrated.

Referee Report

4 major / 5 minor

Summary. The paper proposes HARIS, a hierarchical two-agent framework for multi-hop claim verification. A high-level reasoning agent constructs a verification chain and emits factual questions, while a low-level search agent performs iterative retrieval and reports evidence; both agents are trained with GRPO in an alternating, coordinated scheme, with the search agent rewarded by an LLM-as-a-Judge against GPT-4o-generated pseudo ground-truth answers. The authors evaluate on EX-FEVER, HOVER, and CHECKWHY, reporting that HARIS, using a Qwen3-4B backbone, outperforms RAG, decomposition, program-guided, FOL, and search-augmented baselines, including a 32B Search-o1 baseline, and they provide ablations on single-agent design, coordination rounds, and reward functions.

Significance. If the reported results are reliable, HARIS is a meaningful contribution: it explicitly models the interleaving of reasoning and search, trains both roles with outcome-based RL, and provides a human-evaluated LLM-as-a-Judge reward for search-agent training. The paper also releases code, which supports reproducibility. The conceptual separation of reasoning and search is well motivated, and the qualitative examples illustrate the intended behavior. However, the central 'consistently outperforms all baselines' claim is currently supported by single-run, 500-instance evaluations without statistical testing or error bars, and the sole same-backbone comparison contains a concrete table error. The contribution is therefore promising but not yet fully evidenced as presented.

major comments (4)

[Section B.1, Table 5] In Table 5, the FOLK row reports F1 and accuracy values that are identical to the HARIS row for all five dataset/hop settings (e.g., HOVER 2-hops F1=69.31, Acc=69.40; EX-FEVER 2-hops F1=80.12, Acc=80.20). This cannot be a genuine FOLK run and appears to be a copy/paste error. Since Table 5 is the only direct same-backbone comparison and Section B.1 explicitly claims HARIS significantly outperforms baselines on this table, the controlled comparison is invalid as printed. The FOLK row must be corrected or removed, and the authors should verify that no other rows in Table 5 or elsewhere are affected by the same issue.
[Section 5.1, Table 1] The headline claim that HARIS 'consistently outperforms all baseline methods' is not supported by the reported statistics. Evaluation is on 500 stratified test instances per setting (Section 4.1), and no table reports variance, confidence intervals, multiple seeds, or significance tests. For a binary accuracy near 0.6, the standard error of the difference between two methods on the same 500 instances is roughly 3.1 percentage points, so a 95% interval spans about ±6 points. Several headline margins are within this range: HOVER accuracy over QACheck is +1.8, +2.2, and +2.0 for 2-, 3-, and 4-hop, and HOVER 2-hop accuracy over Search-o1 is only +0.4. The consistency across settings is suggestive, but the paper needs either error bars or significance tests across repeated runs/seeds to substantiate 'consistently outperforms'.
[Section 5.2, Table 3] The text states that increasing coordination rounds N 'generally leads to stronger performance,' but Table 3 shows decreases from N=1 to N=3 for HOVER 2-hops (F1 69.37 to 69.31; Acc 69.40 to 69.40) and HOVER 4-hops (F1 60.87 to 59.84; Acc 61.80 to 61.00). The claim is therefore only partially supported, and the table as presented does not demonstrate a monotonic or consistent benefit of more coordination rounds. The authors should either revise the claim to acknowledge the mixed results, provide additional N values with error bars, or analyze why some settings degrade with additional coordination rounds.
[Section 3.3.1, Appendix A.4] The search-agent training signal depends on GPT-4o pseudo ground-truth answers paired with questions sampled from the reasoning agent's rollouts. The paper does not evaluate the quality of these pseudo-answers; the human evaluation in Appendix D assesses only the LLM-as-a-Judge reward, not the pseudo ground-truth generation. Since noisy pseudo-answers could cause the search agent to retrieve the wrong kind of information and alternately propagate errors into the reasoning agent, the authors should either provide a human evaluation of the synthesized QA pairs, report agreement between GPT-4o pseudo-answers and gold evidence, or include a robustness experiment using gold-derived answers where available.

minor comments (5)

[Appendix C] Sections C.2 and C.3 have identical headings and content: both are titled 'Synthesized Training Data' and present the same Table 11. One of these sections should be removed.
[Figure 4 caption] The caption reads 'Comparision of calls to search agent'; this should be corrected to 'Comparison'.
[Section 6 and Figure 7] The text in Section 6 refers to 'another example (Figure 7)' as part of the case study, but Figure 7 appears in the appendix and is labeled as a different rollout example. The figure numbering and cross-references should be cleaned up so the reader can follow which example is being discussed.
[Section 5.2, LLM-as-a-Judge vs F1] The sentence 'On average it improves over 3% performance' is vague; specify the exact metric (F1, accuracy, or both) and the settings averaged over Table 4.
[Section 4.3] The paper states that all baselines except Search-o1 use GPT-4o, while Search-o1 uses QwQ-32B-preview, and HARIS uses Qwen3-4B. This is potentially a strength, but it should be acknowledged in the main text that API-based baselines with GPT-4o are not frozen-seed deterministic, which further motivates the need for repeated evaluation runs.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: HARIS is evaluated on held-out external benchmarks, its final reward uses the true label, and the self-referential GPT-4o-based search-agent training signal is mitigated by human evaluation and does not drive the headline comparison.

full rationale

The paper's central claim—that HARIS outperforms baselines on EX-FEVER, HOVER, and CHECKWHY—rests on held-out test subsets (Section 4.1: 500 stratified test instances per hop setting) and on an outcome-based reward computed from the true verification label (Eq. 1). The reasoning agent is not trained to match any fitted target; it is rewarded directly for correct Support/Refute predictions. The search agent is trained with a more indirect signal: questions sampled from reasoning-agent rollouts are paired with GPT-4o-generated pseudo-answers, and the search-agent reward is an LLM-as-a-Judge comparison against those pseudo-answers (Eqs. 2–3). This is a partially self-referential training loop, since GPT-4o generates the target and GPT-4o-mini judges the search output, but it is not the basis of the headline result. The final evaluation uses real dataset labels, and the judge itself is validated against two human annotators with Cohen's Kappa 0.81 and 93.3% agreement (Appendix D), so the loop is not an unverified self-confirmation. Author self-citations (Hu et al., 2025a,b) appear only as background context in Section 2.1 and are not load-bearing; no uniqueness theorem or ansatz is imported from the authors' prior work. One serious but non-circularity issue should be flagged: Table 5 lists FOLK Qwen3-4B results identical to HARIS for all five reported rows, which cannot be a genuine independent run; this invalidates the same-backbone comparison as printed and is an evidence-quality concern, not a circular derivation. Likewise, the absence of variance, confidence intervals, and significance tests over 500-instance runs weakens the strength of the 'consistently outperforms' phrasing, but it does not make the derivation circular. Overall, the derivation chain is self-contained against external benchmarks, and there is no step in which a prediction reduces by construction to a fitted input or to a self-citation chain.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central empirical claim rests on the synthetic QA training pipeline and the LLM-as-a-Judge reward. These are domain assumptions rather than formal axioms, and the paper provides a human agreement check for the judge. No new physical or mathematical entities are introduced; the 'agents' are software components, not hypothesized entities.

free parameters (5)

Reward shaping constants = 1.0 / 0.1 / 0.0
Hand-chosen reward values in Eq. 1 and Eq. 3; no sensitivity analysis, but they mainly affect training stability rather than the central empirical claim.
Number of coordination rounds N = 3 (default)
Table 3 compares N=1 and N=3; results are mixed, with HOVER 2-hop and 4-hop slightly worse at N=3, yet the paper claims increasing N generally helps.
Top-k retrieval = 3
Retrieval depth chosen for all methods; not swept.
Training set size = 7,200 examples
Sampled from EX-FEVER and HOVER training data; chosen for compute budget.
Data filter threshold = ~10% removed when GPT-4o outputs 'none'
Post-hoc filtering of synthetic QA pairs in Section 3.3.1.

assumptions (6)

domain assumption GPT-4o pseudo ground-truth answers are accurate enough to train the search agent
Section 3.3.1 and Appendix A.4 use GPT-4o generated answers as targets; no manual verification of the synthesized QA set.
domain assumption LLM-as-a-Judge (GPT-4o-mini) reward correlates with useful information gathering
Section 3.3.3; human eval on 150 samples gives Kappa 0.81, but judge is used as reward throughout training.
domain assumption The FlashRAG Wikipedia corpus contains sufficient evidence for all test claims
Section A.2; retrieval is limited to this corpus, so claims requiring external sources cannot be verified.
domain assumption Binary Support/Refute setup is adequate
Stated in Limitations; consistent with baselines, but excludes Neutral/Not Enough Info classes.
standard math GRPO with outcome-based reward can train question generation and search behavior
Used in Search-R1 and R1-Searcher; the paper relies on prior work for convergence and stability.
domain assumption The 500-sample stratified test subset represents the full test set
Section 4.1 uses stratified sampling of 500 instances per hop setting following Wang and Shu (2023); no comparison to full test set numbers is provided.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Coordinating Search-Informed Reasoning and Reasoning-Guided Search in Claim Verification." pith.science (2026). https://pith.science/paper/CXD3EBSE

@misc{pith2026250607528,
  author       = {Pith},
  title        = {Pith review of: Coordinating Search-Informed Reasoning and Reasoning-Guided Search in Claim Verification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CXD3EBSE}},
  note         = {Machine review of arXiv:2506.07528}
}

read the original abstract

Multi-hop claim verification is inherently challenging, requiring multi-step reasoning to construct verification chains while iteratively searching for information to uncover hidden bridging facts. This process is fundamentally interleaved, as effective reasoning relies on dynamically retrieved evidence, while effective search demands reasoning to refine queries based on partial information. To achieve this, we propose Hierarchical Agent Reasoning and Information Search (HARIS), explicitly modeling the coordinated process of reasoning-driven searching and search-informed reasoning. HARIS consists of a high-level reasoning agent that focuses on constructing the main verification chain, generating factual questions when more information is needed, and a low-level search agent that iteratively retrieves more information, refining its search based on intermediate findings. This design allows each agent to specialize in its respective task, enhancing verification accuracy and interpretability. HARIS is trained using reinforcement learning with outcome-based rewards. Experimental results on the EX-FEVER and HOVER benchmarks demonstrate that HARIS achieves strong performance, greatly advancing multi-hop claim verification.

Figures

Figures reproduced from arXiv: 2506.07528 by the authors.

**Figure 2.** Interaction workflow (left) and coordinated agent training process (right). The reasoning agent constructs [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Demonstrations of GRPO for the reasoning and search agents in HARIS. During reasoning agent training, the policy LLM interacts with the search agent, while for search agent training, it interacts with the search engine. LGRPO(θ) = − 1 G XG i=1 X |oi| t=1 πθ(oi,t | q, oi,<t ) πθold (oi,t | q, oi,<t ) Aˆi,t −βDKL[πθ ∥ πθold ] , (4) where G is the number of groups, Aˆ i,t is the normalized advantage within the grou… view at source ↗

Figures from the paper (7 more)

**Figure 4.** Figure 4: Comparision of calls to search agent during [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

**Figure 5.** Figure 5: Comparison of high-level rewards during rea [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

**Figure 6.** Figure 6: Example of reasoning agent and search agent rollout interaction for a complex multi-hop claim. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

**Figure 7.** Figure 7: Different rollout example from reasoning agent for a complex multi-hop claim. [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]

**Figure 8.** Figure 8: Example of a trained single agent rollout. The input claim ‘Sam Stockman stars in a series that depicted a [PITH_FULL_IMAGE:figures/full_fig_p020_8.png]

**Figure 9.** Figure 9: Example of HARIS’s reasoning agent rollout. The input claim ‘Sam Stockman stars in a series that [PITH_FULL_IMAGE:figures/full_fig_p021_9.png]

**Figure 10.** Figure 10: Preview of the human evaluation panel. 22 [PITH_FULL_IMAGE:figures/full_fig_p022_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

47 extracted references · 28 canonical work pages

[1]

Pepa Atanasova, Jakob Grue Simonsen, Christina Lioma, and Isabelle Augenstein. 2022. Fact checking with insufficient evidence. Transactions of the Association for Computational Linguistics, 10:746--763

work page 2022
[2]

Jifan Chen, Grace Kim, Aniruddh Sriram, Greg Durrett, and Eunsol Choi. 2024. https://aclanthology.org/2024.naacl-long.196/ Complex claim verification with evidence retrieved in the wild . In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)...

work page 2024
[3]

Mingyang Chen, Tianpeng Li, Haoze Sun, Yijie Zhou, Chenzheng Zhu, Haofen Wang, Jeff Z Pan, Wen Zhang, Huajun Chen, Fan Yang, and 1 others. 2025. Research: Learning to reason with search for llms via reinforcement learning. arXiv preprint arXiv:2503.19470

arXiv 2025
[4]

Yunfan Gao, Yun Xiong, Yijie Zhong, Yuxi Bi, Ming Xue, and Haofen Wang. 2025. Synergizing rag and reasoning: A systematic review. arXiv preprint arXiv:2504.15909

arXiv 2025
[5]

Xinyan Guan, Jiali Zeng, Fandong Meng, Chunlei Xin, Yaojie Lu, Hongyu Lin, Xianpei Han, Le Sun, and Jie Zhou. 2025. Deeprag: Thinking to retrieval step by step for large language models. arXiv preprint arXiv:2502.01142

arXiv 2025
[6]

Zhijiang Guo, Michael Schlichtkrull, and Andreas Vlachos. 2022. A survey on automated fact-checking. Transactions of the Association for Computational Linguistics, 10:178--206

work page 2022
[7]

Qisheng Hu, Quanyu Long, and Wenya Wang. 2025 a . Boost: Bootstrapping strategy-driven reasoning programs for program-guided fact-checking. arXiv preprint arXiv:2504.02467

work page Pith review arXiv 2025
[8]

Qisheng Hu, Quanyu Long, and Wenya Wang. 2025 b . https://aclanthology.org/2025.naacl-long.320/ Decomposition dilemmas: Does claim decomposition boost or burden fact-checking performance? In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Lon...

work page 2025

Show all 47 references

[9]

Hyewon Jeon and Jay-Yoon Lee. 2025. Graphcheck: Multi-path fact-checking with entity-relationship graphs. arXiv preprint arXiv:2502.20785

2025
[10]

Yichen Jiang, Shikha Bordia, Zheng Zhong, Charles Dognin, Maneesh Singh, and Mohit Bansal. 2020. https://aclanthology.org/2020.findings-emnlp.309/ H o V er: A dataset for many-hop fact extraction and claim verification . In Findings of the Association for Computational Linguis...

2020
[11]

Zhengping Jiang, Jingyu Zhang, Nathaniel Weir, Seth Ebner, Miriam Wanner, Kate Sanders, Daniel Khashabi, Anqi Liu, and Benjamin Van Durme. 2024. Core: Robust factual precision scoring with informative sub-claim identification. arXiv preprint arXiv:2407.03572

2024 arXiv
[12]

Bowen Jin, Hansi Zeng, Zhenrui Yue, Jinsung Yoon, Sercan Arik, Dong Wang, Hamed Zamani, and Jiawei Han. 2025. Search-r1: Training llms to reason and leverage search engines with reinforcement learning. arXiv preprint arXiv:2503.09516

2025 arXiv
[13]

Jiajie Jin, Yutao Zhu, Xinyu Yang, Chenghao Zhang, and Zhicheng Dou. 2024. https://doi.org/10.48550/ARXIV.2405.13576 Flashrag: A modular toolkit for efficient retrieval-augmented generation research . CoRR, abs/2405.13576

2024 doi arXiv
[14]

Ryo Kamoi, Tanya Goyal, Juan Diego Rodriguez, and Greg Durrett. 2023. https://aclanthology.org/2023.emnlp-main.470 W i CE : Real-world entailment for claims in W ikipedia . In Proceedings of EMNLP

2023
[15]

Joshi, Hanna Moazam, Heather Miller, Matei Zaharia, and Christopher Potts

Omar Khattab, Arnav Singhvi, Paridhi Maheshwari, Zhiyuan Zhang, Keshav Santhanam, Sri Vardhamanan, Saiful Haq, Ashutosh Sharma, Thomas T. Joshi, Hanna Moazam, Heather Miller, Matei Zaharia, and Christopher Potts. 2024. Dspy: Compiling declarative language model calls into self...

2024
[16]

Gonzalez, Hao Zhang, and Ion Stoica

Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph E. Gonzalez, Hao Zhang, and Ion Stoica. 2023. Efficient memory management for large language model serving with pagedattention. In Proceedings of the ACM SIGOPS 29th Symposium on Operating S...

2023
[17]

Xiaoxi Li, Guanting Dong, Jiajie Jin, Yuyao Zhang, Yujia Zhou, Yutao Zhu, Peitian Zhang, and Zhicheng Dou. 2025. Search-o1: Agentic search-enhanced large reasoning models. arXiv preprint arXiv:2501.05366

2025 arXiv
[18]

Yuxuan Liu, Hongda Sun, Wenya Guo, Xinyan Xiao, Cunli Mao, Zhengtao Yu, and Rui Yan. 2025. Bidev: Bilateral defusing verification for complex claim fact-checking. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 541--549

2025
[19]

Yining Lu, Noah Ziems, Hy Dang, and Meng Jiang. 2025. Optimizing decomposition for optimal claim verification. arXiv preprint arXiv:2503.15354

2025 arXiv
[20]

Huanhuan Ma, Weizhi Xu, Yifan Wei, Liuji Chen, Liang Wang, Qiang Liu, Shu Wu, and Liang Wang. 2024. https://aclanthology.org/2024.findings-acl.556/ EX - FEVER : A dataset for multi-hop explainable fact verification . In Findings of the Association for Computational Linguistics...

2024
[21]

Jiatong Ma, Linmei Hu, Rang Li, and Wenbo Fu. 2025. Local: Logical and causal fact-checking with llm-based multi-agents. In Proceedings of the ACM on Web Conference 2025, pages 1614--1625

2025
[22]

Niklas Muennighoff, Nouamane Tazi, Loic Magne, and Nils Reimers. 2023. https://aclanthology.org/2023.eacl-main.148 MTEB : Massive text embedding benchmark . In Proceedings of EACL, pages 2014--2037

2023
[23]

OpenAI. 2024 a . https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/ Gpt-4o mini: advancing cost-efficient intelligence . OpenAI Blog

2024
[24]

OpenAI. 2024 b . https://openai.com/index/hello-gpt-4o/ Hello gpt-4o . OpenAI Blog

2024
[25]

Liangming Pan, Xinyuan Lu, Min-Yen Kan, and Preslav Nakov. 2023 a . https://aclanthology.org/2023.emnlp-demo.23/ QAC heck: A demonstration system for question-guided multi-hop fact-checking . In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Proces...

2023
[26]

Liangming Pan, Xiaobao Wu, Xinyuan Lu, Anh Tuan Luu, William Yang Wang, Min-Yen Kan, and Preslav Nakov. 2023 b . https://aclanthology.org/2023.acl-long.386/ Fact-checking complex claims with program-guided reasoning . In Proceedings of the 61st Annual Meeting of the Associatio...

2023
[27]

Hoang Pham, Thanh-Do Nguyen, and Khac-Hoai Nam Bui. 2025. Verify-in-the-graph: Entity disambiguation enhancement for complex claim verification with interactive graph representation. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Associatio...

2025
[28]

Cheng Qian, Emre Can Acikgoz, Qi He, Hongru Wang, Xiusi Chen, Dilek Hakkani-T \"u r, Gokhan Tur, and Heng Ji. 2025. Toolrl: Reward is all tool learning needs. arXiv preprint arXiv:2504.13958

2025 arXiv
[29]

Qwen. 2024. https://qwenlm.github.io/blog/qwq-32b-preview/ Qwq: Reflect deeply on the boundaries of the unknown

2024
[30]

Qwen. 2025. Qwen3 technical report. arXiv preprint arXiv:2505.09388

2025 arXiv
[31]

Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Y Wu, and 1 others. 2024. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. arXiv preprint arXiv:2402.03300

2024 arXiv
[32]

Guangming Sheng, Chi Zhang, Zilingfeng Ye, Xibin Wu, Wang Zhang, Ru Zhang, Yanghua Peng, Haibin Lin, and Chuan Wu. 2024. Hybridflow: A flexible and efficient rlhf framework. arXiv preprint arXiv: 2409.19256

2024 arXiv
[33]

Jiasheng Si, Yibo Zhao, Yingjie Zhu, Haiyang Zhu, Wenpeng Lu, and Deyu Zhou. 2024. https://aclanthology.org/2024.acl-long.835/ CHECKWHY : Causal fact verification via argument structure . In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistic...

2024
[34]

Huatong Song, Jinhao Jiang, Yingqian Min, Jie Chen, Zhipeng Chen, Wayne Xin Zhao, Lei Fang, and Ji-Rong Wen. 2025. R1-searcher: Incentivizing the search capability in llms via reinforcement learning. arXiv preprint arXiv:2503.05592

2025 arXiv
[35]

Hao Sun, Zile Qiao, Jiayan Guo, Xuanbo Fan, Yingyan Hou, Yong Jiang, Pengjun Xie, Fei Huang, and Yan Zhang. 2025. Zerosearch: Incentivize the search capability of llms without searching. arXiv preprint arXiv:2505.04588

2025 arXiv
[36]

Ziyu Wan, Yunxiang Li, Yan Song, Hanjing Wang, Linyi Yang, Mark Schmidt, Jun Wang, Weinan Zhang, Shuyue Hu, and Ying Wen. 2025. Rema: Learning to meta-think for llms with multi-agent reinforcement learning. arXiv preprint arXiv:2503.09501

2025 arXiv
[37]

Haoran Wang and Kai Shu. 2023. https://aclanthology.org/2023.findings-emnlp.416/ Explainable claim verification via knowledge-grounded reasoning with large language models . In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 6288--6304

2023
[38]

Liang Wang, Nan Yang, Xiaolong Huang, Linjun Yang, Rangan Majumder, and Furu Wei. 2024. Multilingual e5 text embeddings: A technical report. arXiv preprint arXiv:2402.05672

2024 arXiv
[39]

Miriam Wanner, Seth Ebner, Zhengping Jiang, Mark Dredze, and Benjamin Van Durme. 2024 a . https://aclanthology.org/2024.starsem-1.13 A closer look at claim decomposition . In Proceedings of the 13th Joint Conference on Lexical and Computational Semantics (*SEM 2024), pages 153--175

2024
[40]

Miriam Wanner, Benjamin Van Durme, and Mark Dredze. 2024 b . Dndscore: Decontextualization and decomposition for factuality verification in long-form text generation. arXiv preprint arXiv:2412.13175

2024 arXiv
[41]

Junde Wu, Jiayuan Zhu, and Yuyuan Liu. 2025. Agentic reasoning: Reasoning llms with tools for the deep research. arXiv preprint arXiv:2502.04644

2025 arXiv
[42]

Guangzhi Xiong, Qiao Jin, Xiao Wang, Yin Fang, Haolin Liu, Yifan Yang, Fangyuan Chen, Zhixing Song, Dengyu Wang, Minjia Zhang, and 1 others. 2025. Rag-gym: Optimizing reasoning and search agents with process supervision. arXiv preprint arXiv:2502.13957

2025 arXiv
[43]

Fengzhu Zeng and Wei Gao. 2024. J usti LM : Few-shot justification generation for explainable fact-checking of real-world claims. Transactions of the Association for Computational Linguistics, pages 334--354

2024
[44]

Xiaoyan Zhao, Lingzhi Wang, Zhanghao Wang, Hong Cheng, Rui Zhang, and Kam-Fai Wong. 2024. Pacar: Automated fact-checking with planning and customized action reasoning using large language models. In Proceedings of the 2024 Joint International Conference on Computational Lingui...

2024
[45]

Liwen Zheng, Chaozhuo Li, Litian Zhang, Haoran Jia, Senzhang Wang, Zheng Liu, and Xi Zhang. 2025. Mrr-fv: Unlocking complex fact verification with multi-hop retrieval and reasoning. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 26066--26074

2025
[46]

online" 'onlinestring :=

ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...
[47]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Pepa Atanasova, Jakob Grue Simonsen, Christina Lioma, and Isabelle Augenstein. 2022. Fact checking with insufficient evidence. Transactions of the Association for Computational Linguistics, 10:746--763

work page 2022

[2] [2]

Jifan Chen, Grace Kim, Aniruddh Sriram, Greg Durrett, and Eunsol Choi. 2024. https://aclanthology.org/2024.naacl-long.196/ Complex claim verification with evidence retrieved in the wild . In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)...

work page 2024

[3] [3]

Mingyang Chen, Tianpeng Li, Haoze Sun, Yijie Zhou, Chenzheng Zhu, Haofen Wang, Jeff Z Pan, Wen Zhang, Huajun Chen, Fan Yang, and 1 others. 2025. Research: Learning to reason with search for llms via reinforcement learning. arXiv preprint arXiv:2503.19470

arXiv 2025

[4] [4]

Yunfan Gao, Yun Xiong, Yijie Zhong, Yuxi Bi, Ming Xue, and Haofen Wang. 2025. Synergizing rag and reasoning: A systematic review. arXiv preprint arXiv:2504.15909

arXiv 2025

[5] [5]

Xinyan Guan, Jiali Zeng, Fandong Meng, Chunlei Xin, Yaojie Lu, Hongyu Lin, Xianpei Han, Le Sun, and Jie Zhou. 2025. Deeprag: Thinking to retrieval step by step for large language models. arXiv preprint arXiv:2502.01142

arXiv 2025

[6] [6]

Zhijiang Guo, Michael Schlichtkrull, and Andreas Vlachos. 2022. A survey on automated fact-checking. Transactions of the Association for Computational Linguistics, 10:178--206

work page 2022

[7] [7]

Qisheng Hu, Quanyu Long, and Wenya Wang. 2025 a . Boost: Bootstrapping strategy-driven reasoning programs for program-guided fact-checking. arXiv preprint arXiv:2504.02467

work page Pith review arXiv 2025

[8] [8]

Qisheng Hu, Quanyu Long, and Wenya Wang. 2025 b . https://aclanthology.org/2025.naacl-long.320/ Decomposition dilemmas: Does claim decomposition boost or burden fact-checking performance? In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Lon...

work page 2025